[SCEV] Try to push op into ZExt: C * zext (A + B) -> zext (A*C + B*C) #155300
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
llvmbot
added
llvm:analysis
Member
|
@llvm/pr-subscribers-llvm-analysis Author: Florian Hahn (fhahn) ChangesTry to push constant multiply operand into a ZExt containing an add, if possible. In general we are trying to push down ops through ZExt if possible. This is similar to #151227 which did the same for additions. For now this is restricted to adds with a constant operand, which is similar to some of the logic above. This enables some additional simplifications. Alive2 Proof: https://alive2.llvm.org/ce/z/97pbSL Full diff: https://github.com/llvm/llvm-project/pull/155300.diff 6 Files Affected:
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index f60a1e9f22704..40e263efecec6 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -1818,7 +1818,7 @@ const SCEV *ScalarEvolution::getZeroExtendExprImpl(const SCEV *Op, Type *Ty,
if (auto *SM = dyn_cast<SCEVMulExpr>(Op)) {
// zext((A * B * ...)<nuw>) --> (zext(A) * zext(B) * ...)<nuw>
- if (SM->hasNoUnsignedWrap()) {
+ if (SM->hasNoUnsignedWrap() /*|| getUnsignedRange(SM->getOperand(0)).unsignedMulMayOverflow(getUnsignedRange()) ==ConstantRange::OverflowResult::NeverOverflows*/) {
// If the multiply does not unsign overflow then we can, by definition,
// commute the zero extension with the multiply operation.
SmallVector<const SCEV *, 4> Ops;
@@ -3199,6 +3199,22 @@ const SCEV *ScalarEvolution::getMulExpr(SmallVectorImpl<const SCEV *> &Ops,
AddRec->getNoWrapFlags(FlagsMask));
}
}
+
+ // Try to push the constant operand into a ZExt: C + zext (A + B) ->
+ // zext (C*A + C*B) if trunc (C) * (A + B) does not unsigned-wrap.
+ const SCEVAddExpr *InnerAdd;
+ if (match(Ops[1], m_scev_ZExt(m_scev_Add(InnerAdd)))) {
+ const SCEV *NarrowC = getTruncateExpr(LHSC, InnerAdd->getType());
+ if (isa<SCEVConstant>(InnerAdd->getOperand(0)) &&
+ getZeroExtendExpr(NarrowC, Ops[1]->getType()) == LHSC &&
+ hasFlags(StrengthenNoWrapFlags(this, scMulExpr, {NarrowC, InnerAdd},
+ SCEV::FlagAnyWrap),
+ SCEV::FlagNUW)) {
+ auto *Res =
+ getMulExpr(NarrowC, InnerAdd, SCEV::FlagAnyWrap, Depth + 1);
+ return getZeroExtendExpr(Res, Ops[1]->getType(), Depth + 1);
+ };
+ }
}
}
diff --git a/llvm/test/Analysis/ScalarEvolution/max-backedge-taken-count-guard-info.ll b/llvm/test/Analysis/ScalarEvolution/max-backedge-taken-count-guard-info.ll
index 9bf2427eddb9c..7cdf3a2d5fd58 100644
--- a/llvm/test/Analysis/ScalarEvolution/max-backedge-taken-count-guard-info.ll
+++ b/llvm/test/Analysis/ScalarEvolution/max-backedge-taken-count-guard-info.ll
@@ -1231,7 +1231,7 @@ define void @optimized_range_check_unsigned3(ptr %pred, i1 %c) {
; CHECK-NEXT: %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%loop> U: [0,3) S: [0,3) Exits: (-1 + %N)<nsw> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %gep = getelementptr inbounds i16, ptr %pred, i32 %iv
-; CHECK-NEXT: --> {%pred,+,2}<nuw><%loop> U: full-set S: full-set Exits: ((2 * (zext i32 (-1 + %N)<nsw> to i64))<nuw><nsw> + %pred) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: --> {%pred,+,2}<nuw><%loop> U: full-set S: full-set Exits: ((zext i32 (-2 + (2 * %N)<nuw><nsw>)<nsw> to i64) + %pred) LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %iv.next = add nuw nsw i32 %iv, 1
; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%loop> U: [1,4) S: [1,4) Exits: %N LoopDispositions: { %loop: Computable }
; CHECK-NEXT: Determining loop execution counts for: @optimized_range_check_unsigned3
diff --git a/llvm/test/Transforms/IndVarSimplify/rewrite-loop-exit-values-phi.ll b/llvm/test/Transforms/IndVarSimplify/rewrite-loop-exit-values-phi.ll
index e1a042747e0ac..84ae79d53e25e 100644
--- a/llvm/test/Transforms/IndVarSimplify/rewrite-loop-exit-values-phi.ll
+++ b/llvm/test/Transforms/IndVarSimplify/rewrite-loop-exit-values-phi.ll
@@ -76,18 +76,18 @@ define i64 @narow_canonical_iv_wide_multiplied_iv(i32 %x, i64 %y, ptr %0) {
; CHECK-LABEL: @narow_canonical_iv_wide_multiplied_iv(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[SMAX:%.*]] = tail call i32 @llvm.smax.i32(i32 [[X:%.*]], i32 1)
-; CHECK-NEXT: [[MUL_Y:%.*]] = shl i64 [[Y:%.*]], 1
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT: [[IV_MUL:%.*]] = phi i64 [ 1, [[ENTRY]] ], [ [[IV_MUL_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT: [[IV_MUL_NEXT]] = add i64 [[IV_MUL]], [[MUL_Y]]
; CHECK-NEXT: call void @foo()
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
; CHECK-NEXT: [[EC:%.*]] = icmp ne i32 [[IV_NEXT]], [[SMAX]]
; CHECK-NEXT: br i1 [[EC]], label [[LOOP]], label [[EXIT:%.*]]
; CHECK: exit:
-; CHECK-NEXT: [[TMP6:%.*]] = phi i64 [ [[IV_MUL_NEXT]], [[LOOP]] ]
+; CHECK-NEXT: [[TMP1:%.*]] = zext nneg i32 [[SMAX]] to i64
+; CHECK-NEXT: [[TMP2:%.*]] = mul i64 [[Y:%.*]], [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 1
+; CHECK-NEXT: [[TMP6:%.*]] = add nuw nsw i64 [[TMP3]], 1
; CHECK-NEXT: ret i64 [[TMP6]]
;
entry:
diff --git a/llvm/test/Transforms/LoopIdiom/X86/memset-size-compute.ll b/llvm/test/Transforms/LoopIdiom/X86/memset-size-compute.ll
index ea2cfe74be264..0123f15334281 100644
--- a/llvm/test/Transforms/LoopIdiom/X86/memset-size-compute.ll
+++ b/llvm/test/Transforms/LoopIdiom/X86/memset-size-compute.ll
@@ -16,10 +16,10 @@ define void @test(ptr %ptr) {
; CHECK-NEXT: [[LIM_0:%.*]] = phi i32 [ 65, [[ENTRY:%.*]] ], [ 1, [[DEAD:%.*]] ]
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[PTR:%.*]], i64 8
; CHECK-NEXT: [[UMAX:%.*]] = call i32 @llvm.umax.i32(i32 [[LIM_0]], i32 2)
-; CHECK-NEXT: [[TMP0:%.*]] = add nsw i32 [[UMAX]], -1
+; CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i32 [[UMAX]], 3
+; CHECK-NEXT: [[TMP0:%.*]] = add nsw i32 [[TMP2]], -8
; CHECK-NEXT: [[TMP1:%.*]] = zext nneg i32 [[TMP0]] to i64
-; CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
-; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 8 [[SCEVGEP]], i8 0, i64 [[TMP2]], i1 false)
+; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 8 [[SCEVGEP]], i8 0, i64 [[TMP1]], i1 false)
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
; CHECK: for.body:
; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], [[FOR_BODY]] ], [ 1, [[FOR_BODY_PREHEADER]] ]
diff --git a/llvm/test/Transforms/LoopIdiom/add-nsw-zext-fold.ll b/llvm/test/Transforms/LoopIdiom/add-nsw-zext-fold.ll
index df32e60d5065a..fc2b5571250ac 100644
--- a/llvm/test/Transforms/LoopIdiom/add-nsw-zext-fold.ll
+++ b/llvm/test/Transforms/LoopIdiom/add-nsw-zext-fold.ll
@@ -9,10 +9,10 @@ define void @fold_add_zext_to_sext(ptr %dst, i1 %start) {
; CHECK-NEXT: [[TMP0:%.*]] = zext i1 [[START]] to i64
; CHECK-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 2
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP1]]
-; CHECK-NEXT: [[TMP2:%.*]] = sub i32 25, [[START_EXT]]
+; CHECK-NEXT: [[TMP4:%.*]] = shl nuw nsw i32 [[START_EXT]], 2
+; CHECK-NEXT: [[TMP2:%.*]] = sub i32 100, [[TMP4]]
; CHECK-NEXT: [[TMP3:%.*]] = zext nneg i32 [[TMP2]] to i64
-; CHECK-NEXT: [[TMP4:%.*]] = shl nuw nsw i64 [[TMP3]], 2
-; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 4 [[SCEVGEP]], i8 0, i64 [[TMP4]], i1 false)
+; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 4 [[SCEVGEP]], i8 0, i64 [[TMP3]], i1 false)
; CHECK-NEXT: br label %[[LOOP:.*]]
; CHECK: [[LOOP]]:
; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[START_EXT]], %[[ENTRY]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
diff --git a/llvm/test/Transforms/LoopVectorize/runtime-check.ll b/llvm/test/Transforms/LoopVectorize/runtime-check.ll
index f4f29689d9cfb..5f92a281ed7c6 100644
--- a/llvm/test/Transforms/LoopVectorize/runtime-check.ll
+++ b/llvm/test/Transforms/LoopVectorize/runtime-check.ll
@@ -492,14 +492,12 @@ define void @test_scev_check_mul_add_expansion(ptr %out, ptr %in, i32 %len, i32
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
; CHECK: vector.memcheck:
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[OUT:%.*]], i64 12
-; CHECK-NEXT: [[TMP1:%.*]] = add nsw i32 [[LEN]], -7
-; CHECK-NEXT: [[TMP2:%.*]] = zext nneg i32 [[TMP1]] to i64
-; CHECK-NEXT: [[TMP3:%.*]] = shl nuw nsw i64 [[TMP2]], 1
+; CHECK-NEXT: [[TMP1:%.*]] = shl nuw i32 [[LEN]], 1
+; CHECK-NEXT: [[TMP3:%.*]] = zext i32 [[TMP1]] to i64
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[OUT]], i64 [[TMP3]]
-; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[TMP4]], i64 14
; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[IN:%.*]], i64 4
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP2]]
-; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[IN]], [[SCEVGEP1]]
+; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[IN]], [[TMP4]]
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
@@ -511,11 +509,11 @@ define void @test_scev_check_mul_add_expansion(ptr %out, ptr %in, i32 %len, i32
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i32 [[INDEX]], 6
; CHECK-NEXT: [[TMP6:%.*]] = sext i32 [[OFFSET_IDX]] to i64
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i16, ptr [[OUT]], i64 [[TMP6]]
-; CHECK-NEXT: store <4 x i16> zeroinitializer, ptr [[TMP7]], align 2, !alias.scope [[META42:![0-9]+]], !noalias [[META45:![0-9]+]]
-; CHECK-NEXT: store i32 0, ptr [[IN]], align 4, !alias.scope [[META45]]
+; CHECK-NEXT: store <4 x i16> zeroinitializer, ptr [[TMP7]], align 2, !alias.scope [[META37:![0-9]+]], !noalias [[META40:![0-9]+]]
+; CHECK-NEXT: store i32 0, ptr [[IN]], align 4, !alias.scope [[META40]]
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP47:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP42:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
@@ -530,7 +528,7 @@ define void @test_scev_check_mul_add_expansion(ptr %out, ptr %in, i32 %len, i32
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
; CHECK-NEXT: store i32 0, ptr [[IN]], align 4
; CHECK-NEXT: [[CMP7_NOT:%.*]] = icmp sgt i32 [[LEN]], [[IV_NEXT]]
-; CHECK-NEXT: br i1 [[CMP7_NOT]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP48:![0-9]+]]
+; CHECK-NEXT: br i1 [[CMP7_NOT]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP43:![0-9]+]]
; CHECK: exit:
; CHECK-NEXT: ret void
;
|
Member
|
@llvm/pr-subscribers-llvm-transforms Author: Florian Hahn (fhahn) ChangesTry to push constant multiply operand into a ZExt containing an add, if possible. In general we are trying to push down ops through ZExt if possible. This is similar to #151227 which did the same for additions. For now this is restricted to adds with a constant operand, which is similar to some of the logic above. This enables some additional simplifications. Alive2 Proof: https://alive2.llvm.org/ce/z/97pbSL Full diff: https://github.com/llvm/llvm-project/pull/155300.diff 6 Files Affected:
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index f60a1e9f22704..40e263efecec6 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -1818,7 +1818,7 @@ const SCEV *ScalarEvolution::getZeroExtendExprImpl(const SCEV *Op, Type *Ty,
if (auto *SM = dyn_cast<SCEVMulExpr>(Op)) {
// zext((A * B * ...)<nuw>) --> (zext(A) * zext(B) * ...)<nuw>
- if (SM->hasNoUnsignedWrap()) {
+ if (SM->hasNoUnsignedWrap() /*|| getUnsignedRange(SM->getOperand(0)).unsignedMulMayOverflow(getUnsignedRange()) ==ConstantRange::OverflowResult::NeverOverflows*/) {
// If the multiply does not unsign overflow then we can, by definition,
// commute the zero extension with the multiply operation.
SmallVector<const SCEV *, 4> Ops;
@@ -3199,6 +3199,22 @@ const SCEV *ScalarEvolution::getMulExpr(SmallVectorImpl<const SCEV *> &Ops,
AddRec->getNoWrapFlags(FlagsMask));
}
}
+
+ // Try to push the constant operand into a ZExt: C + zext (A + B) ->
+ // zext (C*A + C*B) if trunc (C) * (A + B) does not unsigned-wrap.
+ const SCEVAddExpr *InnerAdd;
+ if (match(Ops[1], m_scev_ZExt(m_scev_Add(InnerAdd)))) {
+ const SCEV *NarrowC = getTruncateExpr(LHSC, InnerAdd->getType());
+ if (isa<SCEVConstant>(InnerAdd->getOperand(0)) &&
+ getZeroExtendExpr(NarrowC, Ops[1]->getType()) == LHSC &&
+ hasFlags(StrengthenNoWrapFlags(this, scMulExpr, {NarrowC, InnerAdd},
+ SCEV::FlagAnyWrap),
+ SCEV::FlagNUW)) {
+ auto *Res =
+ getMulExpr(NarrowC, InnerAdd, SCEV::FlagAnyWrap, Depth + 1);
+ return getZeroExtendExpr(Res, Ops[1]->getType(), Depth + 1);
+ };
+ }
}
}
diff --git a/llvm/test/Analysis/ScalarEvolution/max-backedge-taken-count-guard-info.ll b/llvm/test/Analysis/ScalarEvolution/max-backedge-taken-count-guard-info.ll
index 9bf2427eddb9c..7cdf3a2d5fd58 100644
--- a/llvm/test/Analysis/ScalarEvolution/max-backedge-taken-count-guard-info.ll
+++ b/llvm/test/Analysis/ScalarEvolution/max-backedge-taken-count-guard-info.ll
@@ -1231,7 +1231,7 @@ define void @optimized_range_check_unsigned3(ptr %pred, i1 %c) {
; CHECK-NEXT: %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%loop> U: [0,3) S: [0,3) Exits: (-1 + %N)<nsw> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %gep = getelementptr inbounds i16, ptr %pred, i32 %iv
-; CHECK-NEXT: --> {%pred,+,2}<nuw><%loop> U: full-set S: full-set Exits: ((2 * (zext i32 (-1 + %N)<nsw> to i64))<nuw><nsw> + %pred) LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: --> {%pred,+,2}<nuw><%loop> U: full-set S: full-set Exits: ((zext i32 (-2 + (2 * %N)<nuw><nsw>)<nsw> to i64) + %pred) LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %iv.next = add nuw nsw i32 %iv, 1
; CHECK-NEXT: --> {1,+,1}<nuw><nsw><%loop> U: [1,4) S: [1,4) Exits: %N LoopDispositions: { %loop: Computable }
; CHECK-NEXT: Determining loop execution counts for: @optimized_range_check_unsigned3
diff --git a/llvm/test/Transforms/IndVarSimplify/rewrite-loop-exit-values-phi.ll b/llvm/test/Transforms/IndVarSimplify/rewrite-loop-exit-values-phi.ll
index e1a042747e0ac..84ae79d53e25e 100644
--- a/llvm/test/Transforms/IndVarSimplify/rewrite-loop-exit-values-phi.ll
+++ b/llvm/test/Transforms/IndVarSimplify/rewrite-loop-exit-values-phi.ll
@@ -76,18 +76,18 @@ define i64 @narow_canonical_iv_wide_multiplied_iv(i32 %x, i64 %y, ptr %0) {
; CHECK-LABEL: @narow_canonical_iv_wide_multiplied_iv(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[SMAX:%.*]] = tail call i32 @llvm.smax.i32(i32 [[X:%.*]], i32 1)
-; CHECK-NEXT: [[MUL_Y:%.*]] = shl i64 [[Y:%.*]], 1
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[IV:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT: [[IV_MUL:%.*]] = phi i64 [ 1, [[ENTRY]] ], [ [[IV_MUL_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT: [[IV_MUL_NEXT]] = add i64 [[IV_MUL]], [[MUL_Y]]
; CHECK-NEXT: call void @foo()
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
; CHECK-NEXT: [[EC:%.*]] = icmp ne i32 [[IV_NEXT]], [[SMAX]]
; CHECK-NEXT: br i1 [[EC]], label [[LOOP]], label [[EXIT:%.*]]
; CHECK: exit:
-; CHECK-NEXT: [[TMP6:%.*]] = phi i64 [ [[IV_MUL_NEXT]], [[LOOP]] ]
+; CHECK-NEXT: [[TMP1:%.*]] = zext nneg i32 [[SMAX]] to i64
+; CHECK-NEXT: [[TMP2:%.*]] = mul i64 [[Y:%.*]], [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 1
+; CHECK-NEXT: [[TMP6:%.*]] = add nuw nsw i64 [[TMP3]], 1
; CHECK-NEXT: ret i64 [[TMP6]]
;
entry:
diff --git a/llvm/test/Transforms/LoopIdiom/X86/memset-size-compute.ll b/llvm/test/Transforms/LoopIdiom/X86/memset-size-compute.ll
index ea2cfe74be264..0123f15334281 100644
--- a/llvm/test/Transforms/LoopIdiom/X86/memset-size-compute.ll
+++ b/llvm/test/Transforms/LoopIdiom/X86/memset-size-compute.ll
@@ -16,10 +16,10 @@ define void @test(ptr %ptr) {
; CHECK-NEXT: [[LIM_0:%.*]] = phi i32 [ 65, [[ENTRY:%.*]] ], [ 1, [[DEAD:%.*]] ]
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[PTR:%.*]], i64 8
; CHECK-NEXT: [[UMAX:%.*]] = call i32 @llvm.umax.i32(i32 [[LIM_0]], i32 2)
-; CHECK-NEXT: [[TMP0:%.*]] = add nsw i32 [[UMAX]], -1
+; CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i32 [[UMAX]], 3
+; CHECK-NEXT: [[TMP0:%.*]] = add nsw i32 [[TMP2]], -8
; CHECK-NEXT: [[TMP1:%.*]] = zext nneg i32 [[TMP0]] to i64
-; CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 3
-; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 8 [[SCEVGEP]], i8 0, i64 [[TMP2]], i1 false)
+; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 8 [[SCEVGEP]], i8 0, i64 [[TMP1]], i1 false)
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
; CHECK: for.body:
; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], [[FOR_BODY]] ], [ 1, [[FOR_BODY_PREHEADER]] ]
diff --git a/llvm/test/Transforms/LoopIdiom/add-nsw-zext-fold.ll b/llvm/test/Transforms/LoopIdiom/add-nsw-zext-fold.ll
index df32e60d5065a..fc2b5571250ac 100644
--- a/llvm/test/Transforms/LoopIdiom/add-nsw-zext-fold.ll
+++ b/llvm/test/Transforms/LoopIdiom/add-nsw-zext-fold.ll
@@ -9,10 +9,10 @@ define void @fold_add_zext_to_sext(ptr %dst, i1 %start) {
; CHECK-NEXT: [[TMP0:%.*]] = zext i1 [[START]] to i64
; CHECK-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 2
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP1]]
-; CHECK-NEXT: [[TMP2:%.*]] = sub i32 25, [[START_EXT]]
+; CHECK-NEXT: [[TMP4:%.*]] = shl nuw nsw i32 [[START_EXT]], 2
+; CHECK-NEXT: [[TMP2:%.*]] = sub i32 100, [[TMP4]]
; CHECK-NEXT: [[TMP3:%.*]] = zext nneg i32 [[TMP2]] to i64
-; CHECK-NEXT: [[TMP4:%.*]] = shl nuw nsw i64 [[TMP3]], 2
-; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 4 [[SCEVGEP]], i8 0, i64 [[TMP4]], i1 false)
+; CHECK-NEXT: call void @llvm.memset.p0.i64(ptr align 4 [[SCEVGEP]], i8 0, i64 [[TMP3]], i1 false)
; CHECK-NEXT: br label %[[LOOP:.*]]
; CHECK: [[LOOP]]:
; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[START_EXT]], %[[ENTRY]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
diff --git a/llvm/test/Transforms/LoopVectorize/runtime-check.ll b/llvm/test/Transforms/LoopVectorize/runtime-check.ll
index f4f29689d9cfb..5f92a281ed7c6 100644
--- a/llvm/test/Transforms/LoopVectorize/runtime-check.ll
+++ b/llvm/test/Transforms/LoopVectorize/runtime-check.ll
@@ -492,14 +492,12 @@ define void @test_scev_check_mul_add_expansion(ptr %out, ptr %in, i32 %len, i32
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
; CHECK: vector.memcheck:
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[OUT:%.*]], i64 12
-; CHECK-NEXT: [[TMP1:%.*]] = add nsw i32 [[LEN]], -7
-; CHECK-NEXT: [[TMP2:%.*]] = zext nneg i32 [[TMP1]] to i64
-; CHECK-NEXT: [[TMP3:%.*]] = shl nuw nsw i64 [[TMP2]], 1
+; CHECK-NEXT: [[TMP1:%.*]] = shl nuw i32 [[LEN]], 1
+; CHECK-NEXT: [[TMP3:%.*]] = zext i32 [[TMP1]] to i64
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[OUT]], i64 [[TMP3]]
-; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[TMP4]], i64 14
; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[IN:%.*]], i64 4
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[SCEVGEP]], [[SCEVGEP2]]
-; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[IN]], [[SCEVGEP1]]
+; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[IN]], [[TMP4]]
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
@@ -511,11 +509,11 @@ define void @test_scev_check_mul_add_expansion(ptr %out, ptr %in, i32 %len, i32
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i32 [[INDEX]], 6
; CHECK-NEXT: [[TMP6:%.*]] = sext i32 [[OFFSET_IDX]] to i64
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i16, ptr [[OUT]], i64 [[TMP6]]
-; CHECK-NEXT: store <4 x i16> zeroinitializer, ptr [[TMP7]], align 2, !alias.scope [[META42:![0-9]+]], !noalias [[META45:![0-9]+]]
-; CHECK-NEXT: store i32 0, ptr [[IN]], align 4, !alias.scope [[META45]]
+; CHECK-NEXT: store <4 x i16> zeroinitializer, ptr [[TMP7]], align 2, !alias.scope [[META37:![0-9]+]], !noalias [[META40:![0-9]+]]
+; CHECK-NEXT: store i32 0, ptr [[IN]], align 4, !alias.scope [[META40]]
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP47:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP42:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
@@ -530,7 +528,7 @@ define void @test_scev_check_mul_add_expansion(ptr %out, ptr %in, i32 %len, i32
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
; CHECK-NEXT: store i32 0, ptr [[IN]], align 4
; CHECK-NEXT: [[CMP7_NOT:%.*]] = icmp sgt i32 [[LEN]], [[IV_NEXT]]
-; CHECK-NEXT: br i1 [[CMP7_NOT]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP48:![0-9]+]]
+; CHECK-NEXT: br i1 [[CMP7_NOT]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP43:![0-9]+]]
; CHECK: exit:
; CHECK-NEXT: ret void
;
|
nikic
approved these changes
Aug 26, 2025
Contributor
nikic
left a comment
nikic
left a comment
There was a problem hiding this comment.
LGTM assuming no significant compile-time impact.
Contributor
Author
There was a problem hiding this comment.
Woops, thought I removed that already. Stripped now, thanks
Contributor
There was a problem hiding this comment.
Suggested change
| // Try to push the constant operand into a ZExt: C + zext (A + B) -> | |
| // Try to push the constant operand into a ZExt: C * zext (A + B) -> |
Contributor
There was a problem hiding this comment.
May as well pass NUW here, as you already proved it?
Try to push constant multiply operand into a ZExt containing an add, if possible. In general we are trying to push down ops through ZExt if possible. This is similar to llvm#151227 which did the same for additions. For now this is restricted to adds with a constant operand, which is similar to some of the logic above. This enables some additional simplifications. Alive2 Proof: https://alive2.llvm.org/ce/z/97pbSL
fhahn
force-pushed
the
scev-const-mul-sink
branch
from
3e8ded2 to
f1fc8ab
Compare
fhahn
commented
Aug 26, 2025
Contributor
Author
fhahn
left a comment
fhahn
left a comment
There was a problem hiding this comment.
Compile-time results look neutral overall http://llvm-compile-time-tracker.com/compare.php?from=bac8c8784c848f1dd7ddf44a9a463e6c15c6e594&to=89977e0dec139faae854c94fe932b7b768e60537&stat=instructions:u
Contributor
Author
There was a problem hiding this comment.
Woops, thought I removed that already. Stripped now, thanks
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Aug 26, 2025
…(A*C + B*C) (#155300) Try to push constant multiply operand into a ZExt containing an add, if possible. In general we are trying to push down ops through ZExt if possible. This is similar to llvm/llvm-project#151227 which did the same for additions. For now this is restricted to adds with a constant operand, which is similar to some of the logic above. This enables some additional simplifications. Alive2 Proof: https://alive2.llvm.org/ce/z/97pbSL PR: llvm/llvm-project#155300
fhahn
added a commit
to fhahn/llvm-project
that referenced
this pull request
Aug 28, 2025
Remove the fall-back to constant max BTC if the backedge-taken-count cannot be computed. The constant max backedge-taken count is computed considering loop guards, so to avoid regressions we need to apply loop guards as needed. Also remove the special handling for Mul in willNotOverflow, as this should not longer be needed after 9143746 (llvm#155300).
fhahn
added a commit
to fhahn/llvm-project
that referenced
this pull request
Sep 2, 2025
Remove the fall-back to constant max BTC if the backedge-taken-count cannot be computed. The constant max backedge-taken count is computed considering loop guards, so to avoid regressions we need to apply loop guards as needed. Also remove the special handling for Mul in willNotOverflow, as this should not longer be needed after 9143746 (llvm#155300).
fhahn
added a commit
that referenced
this pull request
Sep 2, 2025
#155672) Remove the fall-back to constant max BTC if the backedge-taken-count cannot be computed. The constant max backedge-taken count is computed considering loop guards, so to avoid regressions we need to apply loop guards as needed. Also remove the special handling for Mul in willNotOverflow, as this should not longer be needed after 9143746 (#155300). PR: #155672
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Sep 2, 2025
…cking deref. (#155672) Remove the fall-back to constant max BTC if the backedge-taken-count cannot be computed. The constant max backedge-taken count is computed considering loop guards, so to avoid regressions we need to apply loop guards as needed. Also remove the special handling for Mul in willNotOverflow, as this should not longer be needed after 9143746 (llvm/llvm-project#155300). PR: llvm/llvm-project#155672
fhahn
added a commit
that referenced
this pull request
Sep 3, 2025
…ing deref. (#155672)" This reverts commit f0df1e3. Recommit with extra check for SCEVCouldNotCompute. Test has been added in b169302. Original message: Remove the fall-back to constant max BTC if the backedge-taken-count cannot be computed. The constant max backedge-taken count is computed considering loop guards, so to avoid regressions we need to apply loop guards as needed. Also remove the special handling for Mul in willNotOverflow, as this should not longer be needed after 9143746 (#155300). PR: #155672
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Sep 3, 2025
… when checking deref. (#155672)" This reverts commit f0df1e3. Recommit with extra check for SCEVCouldNotCompute. Test has been added in b169302. Original message: Remove the fall-back to constant max BTC if the backedge-taken-count cannot be computed. The constant max backedge-taken count is computed considering loop guards, so to avoid regressions we need to apply loop guards as needed. Also remove the special handling for Mul in willNotOverflow, as this should not longer be needed after 9143746 (llvm/llvm-project#155300). PR: llvm/llvm-project#155672
fhahn
added a commit
to fhahn/llvm-project
that referenced
this pull request
Nov 4, 2025
…llvm#155300) Try to push constant multiply operand into a ZExt containing an add, if possible. In general we are trying to push down ops through ZExt if possible. This is similar to llvm#151227 which did the same for additions. For now this is restricted to adds with a constant operand, which is similar to some of the logic above. This enables some additional simplifications. Alive2 Proof: https://alive2.llvm.org/ce/z/97pbSL PR: llvm#155300 (cherry-picked from 9143746)
fhahn
added a commit
to fhahn/llvm-project
that referenced
this pull request
Nov 4, 2025
…ing deref. (llvm#155672)" This reverts commit f0df1e3. Recommit with extra check for SCEVCouldNotCompute. Test has been added in b169302. Original message: Remove the fall-back to constant max BTC if the backedge-taken-count cannot be computed. The constant max backedge-taken count is computed considering loop guards, so to avoid regressions we need to apply loop guards as needed. Also remove the special handling for Mul in willNotOverflow, as this should not longer be needed after 9143746 (llvm#155300). PR: llvm#155672 (cherry picked from commit a434a7a)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Try to push constant multiply operand into a ZExt containing an add, if possible. In general we are trying to push down ops through ZExt if possible. This is similar to #151227 which did the same for additions.
For now this is restricted to adds with a constant operand, which is similar to some of the logic above.
This enables some additional simplifications.
Alive2 Proof: https://alive2.llvm.org/ce/z/97pbSL